Identification of Boiling Two-Phase Flow Regimes Based on Two Kinds of Neural Networks

Abstract

In this paper, the boiling phenomena of steam boiler under atmospheric pressure are simulated by using the UDF program of CFD software. Characteristics including pressure, temperature and vapor fraction respectively for bubble, slug and annular flow regimes are extracted as the input characteristic vectors of the BP neural network and Elman neural network for the purpose of identifying the boiling two-phase (vapor/liquid) flow regimes within wall tubes. It reveals that the rate of recognition accuracy of flow regimes with BP neural network is up to 95.24%, as well as 100% with Elman neural network within the groups taken into consider. By analyzing relations between flow regimes, wall temperature and wall heat transfer coefficient, it is found that changes in flow regimes will cause drastic variation in heat transfer coefficient of the wall surface, and the coefficient reduces rapidly as the wall temperature increases and eventually converge to a minimum. It is a very effective method of using numerical simulation to extract the signal under poor experimental conditions, and is good reference for the further research.